On Mon, Dec 28, 2009 at 12:15 PM, Gökhan Sever <gokhansever@gmail.com> wrote:


On Mon, Dec 28, 2009 at 11:16 AM, Gökhan Sever <gokhansever@gmail.com> wrote:


On Mon, Dec 28, 2009 at 11:07 AM, Robert Kern <robert.kern@gmail.com> wrote:
On Mon, Dec 28, 2009 at 11:00, Gökhan Sever <gokhansever@gmail.com> wrote:

> One interesting thing I have noticed while installing the numpy from the
> source is that numpy dependent libraries must be re-installed and this must
> be a clean re-install. For instance I can't import some matplotlib and scipy
> modules without making a fresh installation for these packages. My attempts
> result with a runtime error.

Please, please, always copy-and-paste the traceback when reporting an
error. I know you aren't formally reporting a bug here, but it always
helps.

> Could someone clarify this point? Is this due
> to API change in the numpy core?

Cython/Pyrex code does a runtime check on the struct sizes of types.
We have carefully added a member to the PyArrayDescr struct; i.e. it
shouldn't cause any actual problems, but Cython does the check
anyways. This affects a few modules in scipy, but shouldn't have
affected anything in matplotlib. The traceback may help us identify
the issue you are seeing.


It is too late for the tracebacks. I have already removed the problematic packages and did clean installs. However, next time I will be more careful while reporting such issues. If it helps, in both matplotlib (via ipython -pylab) and scipy.stats import cases the runtime errors was raised due to numpy.core.multiarray module import.

 
--

Robert Kern

"I have come to believe that the whole world is an enigma, a harmless
enigma that is made terrible by our own mad attempt to interpret it as
though it had an underlying truth."
 -- Umberto Eco
_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
http://mail.scipy.org/mailman/listinfo/numpy-discussion



--
Gökhan

Here is another interesting point to consider. To reproduce the runtime error I mentioned previously I downgraded numpy (from the latest check-out installation) following these steps:

svn co http://svn.scipy.org/svn/numpy/branches/1.3.x/ numpy
cd numpy
python setup.py install
Writing /usr/lib/python2.6/site-packages/numpy-1.3.1.dev8031-py2.6.egg-info

I[2]: import matplotlib.pyplot as plt
Segmentation fault

I[3]: from scipy import stats
Segmentation fault

I have installed matplotlib and scipy using the latest numpy dev version. A little later I will downgrade matplotlib and scipy to their previous stable versions, and compile them using numpy 1.3.x. Afterwards I will update numpy and test to see if I can re-produce the runtime error to provide the tracebacks. First let me know if any tracebacks needed for these segfaults or are these known failures?

================================================================================
Platform     : Linux-2.6.29.6-217.2.3.fc11.i686.PAE-i686-with-fedora-11-Leonidas
Python       : ('CPython', 'tags/r26', '66714')
================================================================================


--
Gökhan


Since no one has replied, I tried to reproduce the runtime error but ended up with different errors. Read more for the details:

svn co http://svn.scipy.org/svn/scipy/tags/0.7.1/ scipy
python setup.py install
Writing /usr/lib/python2.6/site-packages/scipy-0.7.1-py2.6.egg-info

svn co https://matplotlib.svn.sourceforge.net/svnroot/matplotlib/tags/v0_99_0/ matplotlib
python setup.py install
Writing /usr/lib/python2.6/site-packages/matplotlib-0.99.0-py2.6.egg-info

install them using

>>> import numpy
>>> numpy.__version__
'1.3.1.dev8031'

scipy import is fine but matplotlib fails with a different import error. I wonder if the buildbots test the source and releases for against different numpy versions or it is just might system acting weird.

>>> from scipy import stats

>>> import matplotlib.pyplot as plt
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/gsever/Desktop/python-repo/matplotlib/lib/matplotlib/pyplot.py", line 6, in <module>
    from matplotlib.figure import Figure, figaspect
  File "/home/gsever/Desktop/python-repo/matplotlib/lib/matplotlib/figure.py", line 17, in <module>
    import artist
  File "/home/gsever/Desktop/python-repo/matplotlib/lib/matplotlib/artist.py", line 5, in <module>
    from transforms import Bbox, IdentityTransform, TransformedBbox, TransformedPath
  File "/home/gsever/Desktop/python-repo/matplotlib/lib/matplotlib/transforms.py", line 34, in <module>
    from matplotlib._path import affine_transform
ImportError: No module named _path

Anyways, back to the main point. First remove the numpy 1.3.x:

[root@ccn site-packages]# rm -rf numpy
[root@ccn site-packages]# rm -rf numpy-1.3.1.dev8031-py2.6.egg-info

and install from the latest trunk:

svn co http://svn.scipy.org/svn/numpy/trunk numpy
python setup.py install
Writing /usr/lib/python2.6/site-packages/numpy-1.5.0.dev8032-py2.6.egg-info

This time scipy asserts:

>>> from scipy import stats
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python2.6/site-packages/scipy/stats/__init__.py", line 7, in <module>
    from stats import *
  File "/usr/lib/python2.6/site-packages/scipy/stats/stats.py", line 203, in <module>
    from morestats import find_repeats #is only reference to scipy.stats
  File "/usr/lib/python2.6/site-packages/scipy/stats/morestats.py", line 7, in <module>
    import distributions
  File "/usr/lib/python2.6/site-packages/scipy/stats/distributions.py", line 27, in <module>
    import vonmises_cython
  File "numpy.pxd", line 30, in scipy.stats.vonmises_cython (scipy/stats/vonmises_cython.c:2939)
ValueError: numpy.dtype does not appear to be the correct type object

import matplotlib.pyplot as plt yields same as previous import attempt. I must have caught an old issue.


--
Gökhan